fix: add rate limit patterns for 'too many tokens' and 'tokens per day' (#39377)

Merged via squash.

Prepared head SHA: 132a457286
Co-authored-by: gambletan <266203672+gambletan@users.noreply.github.com>
Co-authored-by: altaywtf <9790196+altaywtf@users.noreply.github.com>
Reviewed-by: @altaywtf

src/agents/pi-embedded-helpers.isbillingerrormessage.test.ts

@@ -416,12 +416,19 @@ describe("isLikelyContextOverflowError", () => {
       "exceeded your current quota",
       "This request would exceed your account's rate limit",
       "429 Too Many Requests: request exceeds rate limit",
+      "AWS Bedrock: Too many tokens per day. Please try again tomorrow.",
     ];
     for (const sample of samples) {
       expect(isLikelyContextOverflowError(sample)).toBe(false);
     }
   });
 
+  it("keeps too-many-tokens-per-request context overflow errors out of the rate-limit lane", () => {
+    const sample = "Context window exceeded: too many tokens per request.";
+    expect(isLikelyContextOverflowError(sample)).toBe(true);
+    expect(classifyFailoverReason(sample)).toBeNull();
+  });
+
   it("excludes reasoning-required invalid-request errors", () => {
     const samples = [
       "400 Reasoning is mandatory for this endpoint and cannot be disabled.",
@@ -654,6 +661,11 @@ describe("classifyFailoverReason", () => {
       "rate_limit",
     );
   });
+  it("classifies AWS Bedrock too-many-tokens-per-day errors as rate_limit", () => {
+    expect(
+      classifyFailoverReason("AWS Bedrock: Too many tokens per day. Please try again tomorrow."),
+    ).toBe("rate_limit");
+  });
   it("classifies provider high-demand / service-unavailable messages as overloaded", () => {
     expect(
       classifyFailoverReason(

src/agents/pi-embedded-helpers/errors.ts

@@ -122,7 +122,7 @@ const CONTEXT_WINDOW_TOO_SMALL_RE = /context window.*(too small|minimum is)/i;
 const CONTEXT_OVERFLOW_HINT_RE =
   /context.*overflow|context window.*(too (?:large|long)|exceed|over|limit|max(?:imum)?|requested|sent|tokens)|prompt.*(too (?:large|long)|exceed|over|limit|max(?:imum)?)|(?:request|input).*(?:context|window|length|token).*(too (?:large|long)|exceed|over|limit|max(?:imum)?)/i;
 const RATE_LIMIT_HINT_RE =
-  /rate limit|too many requests|requests per (?:minute|hour|day)|quota|throttl|429\b/i;
+  /rate limit|too many requests|requests per (?:minute|hour|day)|quota|throttl|429\b|tokens per day/i;
 
 export function isLikelyContextOverflowError(errorMessage?: string): boolean {
   if (!errorMessage) {

src/agents/pi-embedded-helpers/failover-matches.ts

@@ -14,6 +14,7 @@ const ERROR_PATTERNS = {
     "usage limit",
     /\btpm\b/i,
     "tokens per minute",
+    "tokens per day",
   ],
   overloaded: [
     /overloaded_error|"type"\s*:\s*"overloaded_error"/i,